OpenMathLib/OpenBLAS
1
0
Fork 0
mirror of https://github.com/OpenMathLib/OpenBLAS synced 2026-05-31 00:45:48 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

Fix ARMV9SME target in DYNAMIC_ARCH and add SME query code for MacOS (#5222)

Browse Source 
* Fix ARMV9SME target and add support_sme1 code for MacOS
* make sgemm_direct unconditionally available on all arm64
* build a (dummy) sgemm_direct kernel on all arm64





* Update dynamic_arm64.c
This commit is contained in:
Martin Kroeker
2025-05-10 13:39:32 -07:00
committed by GitHub
parent 2320e0b757
commit 5141a90993
6 changed files with 44 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
driver/others/dynamic_arm64.c
View File

@@ -43,6 +43,14 @@
#include <sys/auxv.h>
#endif
#ifdef __APPLE__
#include <sys/sysctl.h>
int32_t value;
size_t length=sizeof(value);
int64_t value64;
size_t length64=sizeof(value64);
#endif
extern gotoblas_t gotoblas_ARMV8;
#ifdef DYNAMIC_LIST
#ifdef DYN_CORTEXA53
@@ -120,7 +128,7 @@ extern gotoblas_t gotoblas_ARMV9SME;
#else
#define gotoblas_ARMV9SME gotoblas_ARMV8
#endif
#ifdef DYN_CORTEX_A55
#ifdef DYN_CORTEXA55
extern gotoblas_t gotoblas_CORTEXA55;
#else
#define gotoblas_CORTEXA55 gotoblas_ARMV8
@@ -147,17 +155,17 @@ extern gotoblas_t gotoblas_NEOVERSEV1;
extern gotoblas_t gotoblas_NEOVERSEN2;
extern gotoblas_t gotoblas_ARMV8SVE;
extern gotoblas_t gotoblas_A64FX;
#ifndef NO_SME
extern gotoblas_t gotoblas_ARMV9SME;
#else
#define gotoblas_ARMV9SME gotoblas_ARMV8SVE
#endif
#else
#define gotoblas_NEOVERSEV1 gotoblas_ARMV8
#define gotoblas_NEOVERSEN2 gotoblas_ARMV8
#define gotoblas_ARMV8SVE gotoblas_ARMV8
#define gotoblas_A64FX gotoblas_ARMV8
#endif
#ifndef NO_SME
extern gotoblas_t gotoblas_ARMV9SME;
#else
#define gotoblas_ARMV9SME gotoblas_ARMV8SVE
#define gotoblas_ARMV9SME gotoblas_ARMV8
#endif
extern gotoblas_t gotoblas_THUNDERX3T110;
@@ -168,7 +176,7 @@ extern void openblas_warning(int verbose, const char * msg);
#define FALLBACK_VERBOSE 1
#define NEOVERSEN1_FALLBACK "OpenBLAS : Your OS does not support SVE instructions. OpenBLAS is using Neoverse N1 kernels as a fallback, which may give poorer performance.\n"
#define NUM_CORETYPES 18
#define NUM_CORETYPES 19
/*
* In case asm/hwcap.h is outdated on the build system, make sure
@@ -207,6 +215,7 @@ static char *corename[] = {
"cortexa55",
"armv8sve",
"a64fx",
"armv9sme",
"unknown"
};
@@ -229,6 +238,7 @@ char *gotoblas_corename(void) {
if (gotoblas == &gotoblas_CORTEXA55) return corename[15];
if (gotoblas == &gotoblas_ARMV8SVE) return corename[16];
if (gotoblas == &gotoblas_A64FX) return corename[17];
if (gotoblas == &gotoblas_ARMV9SME) return corename[18];
return corename[NUM_CORETYPES];
}
@@ -266,6 +276,7 @@ static gotoblas_t *force_coretype(char *coretype) {
case 15: return (&gotoblas_CORTEXA55);
case 16: return (&gotoblas_ARMV8SVE);
case 17: return (&gotoblas_A64FX);
case 18: return (&gotoblas_ARMV9SME);
}
snprintf(message, 128, "Core not found: %s\n", coretype);
openblas_warning(1, message);
@@ -277,6 +288,11 @@ static gotoblas_t *get_coretype(void) {
char coremsg[128];
#if defined (OS_DARWIN)
//future #if !defined(NO_SME)
// if (support_sme1()) {
// return &gotoblas_ARMV9SME;
// }
// #endif
return &gotoblas_NEOVERSEN1;
#endif
@@ -439,6 +455,7 @@ static gotoblas_t *get_coretype(void) {
}
break;
case 0x61: // Apple
//future if (support_sme1()) return &gotoblas_ARMV9SME;
return &gotoblas_NEOVERSEN1;
break;
default:
@@ -446,8 +463,8 @@ static gotoblas_t *get_coretype(void) {
openblas_warning(1, coremsg);
}
#if !defined(NO_SME) && defined(HWCAP2_SME)
if ((getauxval(AT_HWCAP2) & HWCAP2_SME)) {
#if !defined(NO_SME)
if (support_sme1()) {
return &gotoblas_ARMV9SME;
}
#endif
@@ -511,6 +528,10 @@ int support_sme1(void) {
if(getauxval(AT_HWCAP2) & HWCAP2_SME){
ret = 1;
}
#endif
#if defined(__APPLE__)
sysctlbyname("hw.optional.arm.FEAT_SME",&value64,&length64,NULL,0);
ret = value64;
#endif
return ret;
}